Change-Id: I238d29ba0250224fa593845ae65192653f58faff
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/528156
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This reverts commit d90777ada3.
Reason for revert: relanding with fix to GrBackendTexture
Original change's description:
> Revert "Remove GrBackendFormat's textureType use from isFormatTexturable call."
>
> This reverts commit 832c817bc8.
>
> Reason for revert: uninitialized value in GrBackendTexture
>
> Original change's description:
> > Remove GrBackendFormat's textureType use from isFormatTexturable call.
> >
> > The goal of this change was to remove the use of GrBackendFormat::textureType()
> > from GrCaps::isFormatTexturable call. Instead we will always pass in a
> > GrTextureType into this call.
> >
> > To do this a lot of plumbing of GrTextureType was added to various call
> > sites. However, this CL halts the plubming up at the proxy level where we
> > get it from the GrBackendFormat still. Future CLs will continue removing
> > these call sites and others that use GrBackendFormat::textureType().
> >
> > Bug: skia:12342
> > Change-Id: Ic0f02b9c7f7402405623b8aa31aa32a9a7c22297
> > Reviewed-on: https://skia-review.googlesource.com/c/skia/+/439277
> > Commit-Queue: Greg Daniel <egdaniel@google.com>
> > Reviewed-by: Brian Salomon <bsalomon@google.com>
>
> TBR=egdaniel@google.com,bsalomon@google.com,skcq-be@skia-corp.google.com.iam.gserviceaccount.com
>
> Change-Id: I354bbbf00be7a86c480009f3e7b36a8777a6bf3a
> No-Presubmit: true
> No-Tree-Checks: true
> No-Try: true
> Bug: skia:12342
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/439338
> Reviewed-by: Greg Daniel <egdaniel@google.com>
> Commit-Queue: Greg Daniel <egdaniel@google.com>
# Not skipping CQ checks because this is a reland.
Bug: skia:12342
Change-Id: I151196f149f9e191d2975b8fe81334f4f8720744
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/439339
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This reverts commit 832c817bc8.
Reason for revert: uninitialized value in GrBackendTexture
Original change's description:
> Remove GrBackendFormat's textureType use from isFormatTexturable call.
>
> The goal of this change was to remove the use of GrBackendFormat::textureType()
> from GrCaps::isFormatTexturable call. Instead we will always pass in a
> GrTextureType into this call.
>
> To do this a lot of plumbing of GrTextureType was added to various call
> sites. However, this CL halts the plubming up at the proxy level where we
> get it from the GrBackendFormat still. Future CLs will continue removing
> these call sites and others that use GrBackendFormat::textureType().
>
> Bug: skia:12342
> Change-Id: Ic0f02b9c7f7402405623b8aa31aa32a9a7c22297
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/439277
> Commit-Queue: Greg Daniel <egdaniel@google.com>
> Reviewed-by: Brian Salomon <bsalomon@google.com>
TBR=egdaniel@google.com,bsalomon@google.com,skcq-be@skia-corp.google.com.iam.gserviceaccount.com
Change-Id: I354bbbf00be7a86c480009f3e7b36a8777a6bf3a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:12342
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/439338
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
The goal of this change was to remove the use of GrBackendFormat::textureType()
from GrCaps::isFormatTexturable call. Instead we will always pass in a
GrTextureType into this call.
To do this a lot of plumbing of GrTextureType was added to various call
sites. However, this CL halts the plubming up at the proxy level where we
get it from the GrBackendFormat still. Future CLs will continue removing
these call sites and others that use GrBackendFormat::textureType().
Bug: skia:12342
Change-Id: Ic0f02b9c7f7402405623b8aa31aa32a9a7c22297
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/439277
Commit-Queue: Greg Daniel <egdaniel@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
This reverts commit 0734c6223c.
Reason for revert: This was fixing the symptom and not the disease
Original change's description:
> Clear the tile backend textures when drawing DDLs
>
> It appears that the flutter SKPs don't contain an initial clear so the background can be random noise.
>
> This only appears to be an issue for the Metal backend.
>
> Change-Id: I4f5c823cdb1a0a76c0704d4d48355c3a0ed75d43
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/423316
> Reviewed-by: Michael Ludwig <michaelludwig@google.com>
> Commit-Queue: Robert Phillips <robertphillips@google.com>
TBR=robertphillips@google.com,reed@google.com,michaelludwig@google.com
Change-Id: Iaa52a66248cef2896fc310ee199c3cef7029e532
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/423581
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
It appears that the flutter SKPs don't contain an initial clear so the background can be random noise.
This only appears to be an issue for the Metal backend.
Change-Id: I4f5c823cdb1a0a76c0704d4d48355c3a0ed75d43
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/423316
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
The main change here is just fusing the following three methods into recreateSKP:
deflateSKP
createCallbackContexts
createSKP
All the remaining changes are just API fallout from that.
Bug: skia:11728
Change-Id: Iae2ce65983ad56c8288bdc830e248394a0055bfb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/389925
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
afaik Chrome only tiles in Y which should yield less threaded text blob contention
Change-Id: I9ff484ed2ce66428e2d7ca6188eba79cd99b1547
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/376608
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
- Migrate our code to SkImage::MakePromiseTexture
- Have DDLTileHelper share one SKP and one set of promise images across all tiles.
- Disallow on-the-fly allocation of mips for promise textures.
Bug: skia:10286
Change-Id: Ie35976958454fc520f3c9d860e6285441260c9f7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/291938
Commit-Queue: Adlai Holler <adlai@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
This intended to let us experiment w/ threaded compilation before having
to worry about upstreaming features.
Change-Id: Id9d1807de0fa16475184203d293e00bfaf5fcc01
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/373736
Commit-Queue: Robert Phillips <robertphillips@google.com>
Reviewed-by: Adlai Holler <adlai@google.com>
Also cleanup some of the duplicate code in SkRecords
Bug: skia:7650
Change-Id: I4d3167a892c126c19a54002beab25c9a6c96fa5d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/357000
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Mike Reed <reed@google.com>
This simplifies debugging DDL rendering by allowing arbitrary tiles
to be dropped.
Change-Id: If1341691ef29159a5d48cf1ae40b6f585a7e21a7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/352576
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
With repositionable DDLs the surface proxies they target need
special handling (i.e., their backing resource size cannot
be known ahead of time).
I'll follow this up with a CL that removes the DDLTarget system from
the GrDrawingManager. AFAICT it is not longer needed.
Change-Id: I0d9189b94726fdf356d54c16de32d7e52e0d1451
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/352116
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
When replaying DDLs w/ offsets it is no longer true that the
dimensions of the recording characterization must match those of
the playback characterization.
Change-Id: I3dba1a91af3d58aaf3619274303f8d2b10a76e24
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/344016
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This isn't hooked up anywhere but breaks up the omnibus CL.
Change-Id: I15c200e57450e7cc8ee95a3f7969926d0eb8487f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/333129
Commit-Queue: Robert Phillips <robertphillips@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Remove done and release distinction. Chrome is not using either as
it tracks texture access using other synchronization mechanisms
(semaphores, flush finish procs). Now there is just fulfill and release
where release is called when the texture can be deleted. Also,
release proc can be null.
Simplify texture idle mechanism as the "flushed" state was only used to
implement the old idea of a release proc. The "finished" idle state is
still used to implement the new release proc. Though, it could also be
removed if GrTexture were to be removed for textures returned by fulfill.
Not directly tied to this bug, but a new YUVA factory will be required
and it's good to clean things up first to avoid adding another
instance of the current complexity.
Bug: skia:10632
Change-Id: I4fe3c0af3f5a591506b1b3c736fd3284a38465a6
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/331836
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
This is everything except for literally removing the class.
Change-Id: I2f16caf865d1bcf9c0f267aed73313c0676a73bb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/327222
Commit-Queue: Adlai Holler <adlai@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
We still occasionally downcast, so this is not airtight,
but it (1) allows us to know where we are downcasting and
(2) lets us move away from GrContext (and hopefully remove
it sooner than later.)
All three canaries are currently broken =( so here we go!
Bug: skia:104662
Change-Id: I84efe132574690b62ea512e194e4f9e318e9c050
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/316218
Commit-Queue: Adlai Holler <adlai@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Prior to this CL, if we failed to create the DDL Recorder's target
proxy while creating the target surface we could create an invalid
DDL.
The specific repro case involved too big of an target proxy but
many other scenarios could result in the same behavior.
Bug: 1105903
Change-Id: I519a072600c168aa590fbe920f4029d08fe29e6f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309777
Commit-Queue: Robert Phillips <robertphillips@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Change-Id: Ia2cfbca8982b57399b6681cbb4501c2933ab4df7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/304576
Auto-Submit: Brian Salomon <bsalomon@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Salomon <bsalomon@google.com>
Before I start updating this to support relocatable DDLs I felt I
should clean it up a bit.
Change-Id: I640d15a40164b33c4c2d7378e37d39fe7d3ff313
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/300926
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
External clients will eventually have to call this to get access
to a direct context from an SkCanvas or SkSurface (which will
only have 'recordingContext' accessors).
Bug: skia:10441
Change-Id: I10e34081277b685fa59d03e1fce1887f3524e0fa
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/300178
Reviewed-by: Adlai Holler <adlai@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This will not be landed until chrome CL 2269958 lands.
Bug: skia:10425
Change-Id: I2a5081201ca3faed5232e8540086bd4c6f865767
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/299292
Commit-Queue: Adlai Holler <adlai@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
In order to emulate OOP-R's behavior, GM needs to pass GPU-backed resources to a DDL recorder.
This change allows GMs to create GPU resources first (in onGpuSetup w/ a direct context) and then use them in onDraw (with only a GrRecordingContext).
Change-Id: Ifa3002af73eb9926f653fb4c4bf4542c0749d658
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/294336
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
OOP-R, on the gpu thread, creates the DDL, pre-compiles its shaders, draws it, flushes and then deletes the DDL. This process triggered a bug (cf. https://skia-review.googlesource.com/c/skia/+/292818 and crbug.com/1056730).
Prior to this CL all the programs were compiled and only at the end was any work flushed - thus it was likely that the bound program would be reset to the correct value when rendering.
With this CL, the addition of the flush right before the DDL deletion, makes it more likely that the wrong program will be bound when rendering begins.
Change-Id: I60479bd429e132d8652bbffde6c8b71094be6225
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/292257
Reviewed-by: Adlai Holler <adlai@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
Besides better matching Viz's behavior this also reduces a lot of choppiness in the composition RenderTask DAG.
In the previous approach DDL draws and compositing draws would be interleaved resulting in a lot of render target swaps.
This necessitated some reorganization bc I wanted to reuse PromiseImageCallbackContext to manage the tiles' promiseImages.
Change-Id: I513bf060a69ff2bfe0e7b82ae72f149dfede632e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/285056
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This better matches Chrome's use of DDLs.
With path, image, and text draws stripped out, here is the perf impact of this change:
before CL after CL
w/ DDLs 7.792 1.038
w/o DDLs 0.800 0.876
This perf improvement (in the DDL case) is from backend texture wrapping SkSurfaces being created w/o initialization. The prior method of SkSurface creation was resulting in double clearing of all the surfaces.
This perf improvement won't be seen by Chrome since they've always being using wrapped backend texture SkSurfaces.
TBR=bsalomon@google.com
Bug: 1056730
Change-Id: Ic04d322cad96df845e75437211208495862c6555
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/283866
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This reverts commit 7ae9d2fca6.
Reason for revert: Triggering Vulkan Debug layer errors
Original change's description:
> Update DDL test harness to use backendTextures to back tiles
>
> This better matches Chrome's use of DDLs.
>
> With path, image, and text draws stripped out, here is the perf impact of this change:
>
> before CL after CL
> w/ DDLs 7.792 1.038
> w/o DDLs 0.800 0.876
>
> This perf improvement (in the DDL case) is from backend texture wrapping SkSurfaces being created w/o initialization. The prior method of SkSurface creation was resulting in double clearing of all the surfaces.
>
> This perf improvement won't be seen by Chrome since they've always being using wrapped backend texture SkSurfaces.
>
> TBR=bsalomon@google.com
> Change-Id: Ice3993ca125fce37804e58c353c265cf659dbe2f
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/283456
> Reviewed-by: Brian Salomon <bsalomon@google.com>
> Commit-Queue: Robert Phillips <robertphillips@google.com>
TBR=egdaniel@google.com,bsalomon@google.com,robertphillips@google.com
Change-Id: Ife023ede0774ec2cce4c0d6e7708c036347ebf54
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/283648
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This better matches Chrome's use of DDLs.
With path, image, and text draws stripped out, here is the perf impact of this change:
before CL after CL
w/ DDLs 7.792 1.038
w/o DDLs 0.800 0.876
This perf improvement (in the DDL case) is from backend texture wrapping SkSurfaces being created w/o initialization. The prior method of SkSurface creation was resulting in double clearing of all the surfaces.
This perf improvement won't be seen by Chrome since they've always being using wrapped backend texture SkSurfaces.
TBR=bsalomon@google.com
Change-Id: Ice3993ca125fce37804e58c353c265cf659dbe2f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/283456
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This reverts commit f35dfabe40.
Reason for revert:
Relanding - usage of the flag has been removed from the bots in:
https://skia-review.googlesource.com/c/skia/+/275688 (Remove bot to record DDL creation time)
Original change's description:
> Revert "Update skpbench's DDL timing"
>
> This reverts commit 6f0124ad68.
>
> Reason for revert: need to remove flag on bots first
>
> Original change's description:
> > Update skpbench's DDL timing
> >
> > Chrome is seeing some extra overhead when using DDLs for rasterization. This CL updates skpbench to try to replicate their usage of DDLs (or, at least, better illustrate the overhead of using DDLs).
> >
> > Bug: skia:9455
> > Change-Id: I2abc7cf2d597c97d1d7a47425064c621a7ef0eb3
> > Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275496
> > Reviewed-by: Greg Daniel <egdaniel@google.com>
> > Commit-Queue: Robert Phillips <robertphillips@google.com>
>
> TBR=egdaniel@google.com,robertphillips@google.com
>
> Change-Id: I87b8c78355b81f5f6c90b74f59f176f3cdbdfdaa
> No-Presubmit: true
> No-Tree-Checks: true
> No-Try: true
> Bug: skia:9455
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275681
> Reviewed-by: Robert Phillips <robertphillips@google.com>
> Commit-Queue: Robert Phillips <robertphillips@google.com>
TBR=egdaniel@google.com,robertphillips@google.com
Change-Id: I18446e385edb7827fae89541a03545194fe84f6a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:9455
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275691
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This reverts commit 6f0124ad68.
Reason for revert: need to remove flag on bots first
Original change's description:
> Update skpbench's DDL timing
>
> Chrome is seeing some extra overhead when using DDLs for rasterization. This CL updates skpbench to try to replicate their usage of DDLs (or, at least, better illustrate the overhead of using DDLs).
>
> Bug: skia:9455
> Change-Id: I2abc7cf2d597c97d1d7a47425064c621a7ef0eb3
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275496
> Reviewed-by: Greg Daniel <egdaniel@google.com>
> Commit-Queue: Robert Phillips <robertphillips@google.com>
TBR=egdaniel@google.com,robertphillips@google.com
Change-Id: I87b8c78355b81f5f6c90b74f59f176f3cdbdfdaa
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:9455
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275681
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
Chrome is seeing some extra overhead when using DDLs for rasterization. This CL updates skpbench to try to replicate their usage of DDLs (or, at least, better illustrate the overhead of using DDLs).
Bug: skia:9455
Change-Id: I2abc7cf2d597c97d1d7a47425064c621a7ef0eb3
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275496
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This CL:
1) Fixes a GrTexture access in GrTextureEffect that was blocking pre-compilation
2) Adds program pre-compilation to the DDL Via - which would've caught the GrTextureEffect problem on the bots
3) Adds some #if'ed out code for collecting program pre-compilation stats
Bug: skia:9455
Change-Id: Ibcb07ae855b7a644e1f22c3427a928f116ab300d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275336
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
With the addition of the DDL program pre-compilation we need to know how it is working.
This CL also fixes some threading bugs.
Bug: skia:9455
Change-Id: I20da58a7f1b19685687fae1d159d4e0db8a4964d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/273001
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
This also allows us to remove all the one off Fence code that we
implemented in all the backend TestContexts
Change-Id: I9ff7ba4690cf3f19a180f51fc510991a112bb62c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/272456
Reviewed-by: Chris Dalton <csmartdalton@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
This is through a private API but it, at least, connects things end-to-end.
Bug: skia:9455
Change-Id: Ib34d49c5c4e4cfa5fa599afc5c967fcadc3de10e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/268627
Commit-Queue: Robert Phillips <robertphillips@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>
unused declaration
passing std::move() to const&
These are triggering warnings in google3
Change-Id: I12cebd0a8fd218e7755718fed7acec7908d386a1
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/271138
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Mike Reed <reed@google.com>
TileData now gets a pointer to the final surface
This allows the tile to, once rendered, compose itself into
the final surface
DDLTileHelper now stores the TileData in a dumb array
SkTArray is overkill and, since TileData*s are being doled
out to threads, we never want reallocation
Added DDLTileHelper::kickOffThreadedWork
The old code only performed DDL creation in parallel. This
entry point also replays the DDLs and composes them into the
final surface in parallel
Change-Id: I66e02ef7f8291b4d402e22bee0ad3546e930609e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/270796
Reviewed-by: Greg Daniel <egdaniel@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
